Method and apparatus for removing noise from feature vectors

Abstract

A method and computer-readable medium are provided for identifying clean signal feature vectors from noisy signal feature vectors. The method is based on variational inference techniques. One aspect of the invention includes using an iterative approach to identify the clean signal feature vector. Another aspect of the invention includes using the variance of a set of noise feature vectors and / or channel distortion feature vectors when identifying the clean signal feature vectors. Further aspects of the invention use mixtures of distributions of noise feature vectors and / or channel distortion feature vectors when identifying the clean signal feature vectors. Additional aspects of the invention include using a variance for the noisy signal feature vector conditioned on fixed values of noise, channel transfer function, and clean speech, when identifying the clean signal feature vector.

Description

BACKGROUND OF THE INVENTION[0001]The present invention relates to estimating the feature vectors corresponding to different sources that were combined to produce input feature vectors.[0002]A pattern recognition system, such as a speech recognition system, takes an input signal and attempts to decode the signal to find a pattern represented by the signal. For example, in a speech recognition system, a speech signal is received by the recognition system and is decoded to identify a string of words represented by the speech signal.[0003]To decode the incoming signal, most recognition systems utilize one or more models that describe the likelihood that a portion of the test signal represents a particular pattern. Typically, these models do not operate directly on the incoming signal, but instead operate on a feature vector representation of the incoming signal. In speech recognition, such feature vectors can be produced through techniques such as linear predictive coding (LPC), LPC der...

AI Technical Summary

Problems solved by technology

However, removing noise and channel distortion from the incoming signal itself is computationally difficult because of the large amount of data that has to be processed.

Because the noise and channel distortion vary across the sequence of input features, techniques that use this approximation do not accurately remove the noise or channel distortion.

However, prior art systems do not take the error present in the approximation into account when identifying possible combinations of noise, channel distortion, and original signals based on the incoming signal.

If the form of the approximation is not accurate, the resulting identified combination of noise, channel distortion, and original signal will be inaccurate.

However, the prior art does not provide a means for adjusting the form of the approximation to improve the resulting identified combination.

Images